AW: ECL on very small chips?

Andreas Thiele andreas at atp-media.de
Sun Mar 27 12:23:05 UTC 2016



> -----Ursprüngliche Nachricht-----
> Von: ecl-devel [mailto:ecl-devel-bounces at common-lisp.net] Im Auftrag
> von Pascal J. Bourguignon
> Gesendet: Sonntag, 27. März 2016 00:32
> An: ecl-devel at common-lisp.net
> Betreff: Re: ECL on very small chips?
> 
> "Andreas Thiele" <andreas at atp-media.de>
> writes:
> 
> > can I use ECL to write software for a chip without OS?
> 
> Yes.
> 
> 
> > In my case I’d like to write software for NXP1769 which is ARM Cortex
> > M3, 64kB Ram, 512kB Flash.
> 
> Anything with more than one bit of memory.
> 
> 
> 
> You didn't ask if you could develop software using ECL running on this
> chip.  You can use ECL (or any other CL implementation), at all phases
> of the creation of software for a chip without an OS.
> 
> I would start by writing LAP (Lisp Assembler Program) for that chip.
> Then I would write in lisp an emulator of that chip.  The reason why
> you
> don't want to use the chip itself is that it doesn't have an OS,
> therefore it must be rather hard to debug code you sent there, (even
> with
> a logic analyser).  By writing the emulator in lisp, that means that
> you
> can easily instrumentalize it to help you in debugging.
> 
> Once you have a LAP, you can implement a compiler for a subset of
> Common
> Lisp targetting this LAP.
> 
> Let's remember, you didn't ask to run an interactive lisp with an
> development environment ON this chip.  Therefore you may not need a
> garbage collector, a compiler or an interpreter, strings, cons cells,
> pathnames, multidimensional arrays, CLOS objects, bignums, ratios, etc.
> Probably, for the application you have in mind, you only need
> (signed-byte 32) and vectors, perhaps structures.  You may not need to
> implement dynamic unwinding, the condition system with restarts and so
> on.  After all, all your other colleagues only run C code on those
> chips.  So a small subset of Common Lisp can be all you really need.
> 
> But notice that in the code of your macros, you can use the full Common
> Lisp language, since macros are evaluated at compilation time.  It's in
> the lisp code generated by your macros that you must restrict yourself
> to your CL subset.  (Cf. eg. parenscript).
> 
> You would  define this CL subset also as a package that exports only
> the
> operators included in that subset.  Say the ANDREAS-THIELE-LISP
> package, nickname ATL.
> 
> So you can now develop your software for this chip, as a normal CL
> program using this subset of Common Lisp, that is, using the ATL
> package
> instead of the CL package, on any CL implementation, using all the
> tools
> of your Common Lisp implementation and IDE (choose the one with the
> best
> debugger).
> 
> Once it's good and debugged as a normal CL program, you compile it with
> your compiler instead of using the CL compiler, and you obtain the
> bytes
> to be sent to the chip.  You don't send them to the chip!
> 
> You send them to your emulator, and you test and debug on your emulator
> in Common Lisp, using all the tools of your Common Lisp implementation
> and IDE.
> 
> When it's good and debugged as binary program for you chip, then you
> can
> send it to the chip, and test the final product with whatever tools you
> have there.
> 
> 
> Ok, it may sound complex explained like this, but it's one of the most
> fun way to build a program.
> 
> 
> This is how they wrote Crash Banditcoot, for example.
> https://en.wikipedia.org/wiki/Crash_Bandicoot
> https://en.wikipedia.org/wiki/Game_Oriented_Assembly_Lisp
> http://all-things-andy-gavin.com/2011/02/02/making-crash-bandicoot-
> part-1/
> http://all-things-andy-gavin.com/2011/10/25/lispings-ala-john-mccarthy/
> 
> 
> Finally, remember that the original LISP was written on a machine with
> 36-bit words and 32 Kwords; that's 144 KB. (each 36-bit word could
> store
> a cons cell with two 15-bit pointers and two 3-bit type tags).
> 
> There were lisp implementations on 8-bit processors (eg. on the Apple
> ][
> 6502 with 64 KB of addressing space).
> 
> In the case of CL, the name of the symbols alone take already 12 KB:
> 
> (let ((z 0)) (do-symbols (s "CL" z) (incf z (length (symbol-name s)))))
> --> 11271
> 
> (but that's uncompressed, you can be smart).
> 
> 
> So if your purpose was to implement a Common Lisp system on your chip,
> with 512 KB of flash memory, I'd say that it would be perfectly
> possible, but easier by starting from scratch to take into account the
> size limitations.
> 
> 
> --
> __Pascal Bourguignon__                 http://www.informatimago.com/
> “The factory of the future will have only two employees, a man and a
> dog. The man will be there to feed the dog. The dog will be there to
> keep the man from touching the equipment.” -- Carl Bass CEO Autodesk
> 

Thank you for the inspiring and enlightening answer.

Yes my question was a bit unspecific. The trigger for my question was an unbelievable turnaround time of about 2 minutes (edit, compile, reload, test again) I came across when an embedded box currently under development came onto my desk. This one is written by a junior c programmer with lengthy, sparsely commented c code containing cryptic function names. My job here is programming common lisp, nearly exclusively. So I thought about having a more dynamic development system on the box, which allows dynamic compilation of functions during runtime to not have to reboot the whole box just because one function gets modified. Meanwhile I thought about ECL, PicoLisp, miniPicoLisp, Lua and even Javascript, but I abandoned them all.

Although your emulator suggestion first made me grin, I think it is a possible way to go. Problem is, the chip is connected to other hardware, which has slow communication protocols. Now I think about connecting these hardware components to my emulator (to be written). This could indeed be an approach. I have to think :)

Andreas 

P.S. After browsing the ARM Thumb instruction set I guess writing the emulator will not be trivial :)





More information about the ecl-devel mailing list